1 GWAS Strategy

GWAS was run using MLM model in GCTA1.93.2. Note that I tried different strategies to directly fit covariates or pre-adjust phenotypes by the covariates. The beta correlation between different strategies will be shown for Manhattan Plots as below.

Take SRS_RMB_sum in Probands (with FSIQ included as covariate) for example, I tried:

  • Strategy 1: Phenotype pre-adjusted by age, sex, chip, FSIQ
  • Strategy 2: Phenotype pre-adjusted by age, sex, chip, FSIQ and 20 PCs
  • Strategy 3: directly fit age, sex, chip, FSIQ
  • Strategy 4: directly fit age, sex, chip, FSIQ and 20 PCs
Beta Correlation between Different GWAS Strategies

Beta Correlation between Different GWAS Strategies


Note that considering the GWAS sample size, computational time and false positive rates, we will report the results below:

  • based on Strategy2 for the same phenotype
  • based on sum measurement for the same phenotype


2 Probands

2.1 All Individuals

  • GWAS was run on all individuals including diverse ancestry backgrounds.
  • Signals with association p-value < 1e-5 will be shown for Manhattan Plots.

2.1.1 Association Summary

2.1.1.1 Fitting FSIQ


2.1.1.2 Not fitting FSIQ


2.1.2 QQ Plot

2.1.2.1 Primary Variable

2.1.2.1.1 Fitting FSIQ


2.1.2.1.2 Not fitting FSIQ


2.1.2.2 Secondary Variable

2.1.2.2.1 Fitting FSIQ


2.1.2.2.2 Not fitting FSIQ


2.1.3 Manhattan Plot

2.1.3.1 Primary Variable

2.1.3.1.1 Fitting FSIQ


2.1.3.1.2 Not fitting FSIQ


2.1.3.2 Secondary Variable

2.1.3.2.1 Fitting FSIQ


2.1.3.2.2 Not fitting FSIQ


2.2 Europeans Only

  • 6861972 QCd SNPs with MAF > 0.01 included
  • 1946 European individuals are included
  • Signals with association p-value < 1e-5 will be shown for Manhattan Plots.


2.2.1 Association Summary

2.2.1.1 Fitting FSIQ


2.2.1.2 Not fitting FSIQ


2.2.2 QQ Plot

2.2.2.1 Primary Variable

2.2.2.1.1 Fitting FSIQ


2.2.2.1.2 Not fitting FSIQ


2.2.2.2 Secondary Variable

2.2.2.2.1 Fitting FSIQ


2.2.2.2.2 Not fitting FSIQ


2.2.3 Manhattan Plot

2.2.3.1 Primary Variable

2.2.3.1.1 Fitting FSIQ


2.2.3.1.2 Not fitting FSIQ


2.2.3.2 Secondary Variable

2.2.3.2.1 Fitting FSIQ


2.2.3.2.2 Not fitting FSIQ


3 Probands & Unaffected Siblings

  • Combined phenotypes for Probands and Unaffected Siblings are pre-adjusted by covariates (20 PCs, sex, chip and ASD diagnosis) and then RINT.
  • Based on the phenotype distribution, we run GWAS on combined data.

Note that there is no available information for FSIQ and Age for Unaffected Siblings, thus FSIQ and Age will not be included as covariates for the combined data analysis.

3.1 All Individuals

  • GWAS was run on all individuals including diverse ancestry backgrounds.
  • Signals with association p-value < 1e-5 will be shown for Manhattan Plots.

3.1.1 Association Summary


3.1.2 QQ Plot

3.1.2.1 Primary Variable


3.1.2.2 Secondary Variable


3.1.3 Manhattan Plot

3.1.3.1 Primary Variable


3.1.3.2 Secondary Variable


3.2 Europeans Only

  • GWAS was run on European individuals with N=3544.
  • Signals with association p-value < 1e-5 will be shown for Manhattan Plots for Manhattan Plots.

3.2.1 Association Summary


3.2.2 QQ Plot

3.2.2.1 Primary Variable


3.2.2.2 Secondary Variable

3.2.3 Manhattan Plot

3.2.3.1 Primary Variable


3.2.3.2 Secondary Variable


4 Heritability Estimation

4.1 Probands

  • Unrelated Individuals (GRM < 0.05) were used for performing GREML analysis.
  • Using different models based on phenotypes:
    • Est.1-1: pre-adjust by Age, Sex and Chip
    • Est.1-2: directly fit covariates: Age, Sex and Chip
    • Est.2-1: pre-adjust by Age, Sex, Chip and 20PCs
    • Est.2-2: directly fit covariates: Age, Sex, Chip and 20PCs
    • Est.3-1: pre-adjust by Age, Sex, Chip and FSIQ
    • Est.3-2: directly fit covariates: Age, Sex, Chip and FSIQ
    • Est.4-1: pre-adjust by Age, Sex, Chip, FSIQ and 20PCs
    • Est.4-2: directly fit covariates: Age, Sex, Chip, FSIQ and 20PCs

4.1.1 Primary Variable

4.1.2 Secondary Variable


4.2 Probands & Unaffected Siblings

  • Unrelated Individuals (GRM < 0.05) were used for performing GREML analysis.
  • Using different models based on phenotypes:
    • Est.1-1: pre-adjust by ASD diagnosis, Sex and Chip
    • Est.1-2: directly fit covariates: ASD diagnosis, Sex and Chip
    • Est.2-1: pre-adjust by ASD diagnosis, Sex, Chip and 20PCs
    • Est.2-2: directly fit covariates: ASD diagnosis, Sex, Chip and 20PCs

4.2.1 Primary Variable


4.2.2 Secondary Variable


5 Genetic Correlation

5.1 Probands

5.1.1 RRBs

5.1.2 RRBs VS. Public

5.2 Probands & Unaffected Siblings

5.2.1 RRBs

5.2.2 RRBs VS. Public

[Problems for REML, unrelated EURs for combined data is only ~2K, similar magnitude as probands.]